Vehicle accident image processing method and apparatus

Abstract

N vehicle accident images are obtained, where N is a natural number greater than or equal to 2. N feature vectors are obtained by inputting the vehicle accident images into a trained convolutional neural network, where the N feature vectors respectively correspond to the vehicle accident images. A distance is calculated between any two feature vectors of the N feature vectors. A determination is made that two vehicle accident images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold.

What is claimed is: 
     
       1. A computer-implemented method for vehicle accident image processing, comprising:
 obtaining a group of N vehicle accident images, wherein N is a natural number greater than or equal to 2; 
 obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images into a trained convolutional neural network; 
 calculating respective distances between pairs of feature vectors among the N feature vectors, the pairs of feature vectors corresponding respectively to pairs of images among the N vehicle accident images; 
 determining that a pair of images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold; 
 determining a number of the vehicle accident images that are abnormal, wherein the number of the vehicle accident images that are abnormal is equal to M; 
 calculating a first probability based on M and N; 
 determining that the first probability is greater than a second predetermined threshold; and 
 in response, classifying the group of the N vehicle accident images as abnormal. 
 
     
     
       2. The computer-implemented method of  claim 1 , wherein the trained convolutional neural network is trained using at least one positive sample pair, wherein the training comprises:
 obtaining at least one positive sample pair, wherein each positive sample pair comprises a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and 
 training the convolutional neural network by using the at least one positive sample pair, to decrease a sum of at least one first distance respectively corresponding to the positive sample pair, wherein the at least one first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network. 
 
     
     
       3. The computer-implemented method of  claim 1 , wherein the trained convolutional neural network is trained by at least one negative sample pair, wherein the training comprises:
 obtaining at least one negative sample pair, wherein each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and 
 training the convolutional neural network by using the at least one negative sample pair, to increase a sum of at least one second distance respectively corresponding to the negative sample pair, wherein the at least one second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network. 
 
     
     
       4. The computer-implemented method of  claim 1 , wherein the trained convolutional neural network is trained by at least one positive sample pair and at least one negative sample pair, wherein the training comprises:
 obtaining at least one positive sample pair and at least one negative sample pair, wherein each positive sample pair comprises a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and 
 training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting a sum of at least one second distance corresponding to the negative sample pair from a sum of at least one first distance corresponding to the positive sample pair, wherein the at least one first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the at least one second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network. 
 
     
     
       5. The computer-implemented method of  claim 1 , wherein the distance is a Euclidean distance. 
     
     
       6. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
 obtaining a group of N vehicle accident images, wherein N is a natural number greater than or equal to 2; 
 obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images into a trained convolutional neural network; 
 calculating respective distances between pairs of feature vectors among the N feature vectors, the pairs of feature vectors corresponding respectively to pairs of images among the N vehicle accident images; 
 determining that a pair of images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold; 
 determining a number of the vehicle accident images that are abnormal, wherein the number of the vehicle accident images that are abnormal is equal to M; 
 calculating a first probability based on M and N; 
 determining that the first probability is greater than a second predetermined threshold; and 
 in response, classifying the group of the N vehicle accident images as abnormal. 
 
     
     
       7. The non-transitory, computer-readable medium of  claim 6 , wherein the trained convolutional neural network is trained using at least one positive sample pair, wherein the training comprises:
 obtaining at least one positive sample pair, wherein each positive sample pair comprises a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and 
 training the convolutional neural network by using the at least one positive sample pair, to decrease a sum of at least one first distance respectively corresponding to the positive sample pair, wherein the at least one first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network. 
 
     
     
       8. The non-transitory, computer-readable medium of  claim 6 , wherein the trained convolutional neural network is trained by at least one negative sample pair, wherein the training comprises:
 obtaining at least one negative sample pair, wherein each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and 
 training the convolutional neural network by using the at least one negative sample pair, to increase a sum of at least one second distance respectively corresponding to the negative sample pair, wherein the at least one second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network. 
 
     
     
       9. The non-transitory, computer-readable medium of  claim 6 , wherein the trained convolutional neural network is trained by at least one positive sample pair and at least one negative sample pair, wherein the training comprises:
 obtaining at least one positive sample pair and at least one negative sample pair, wherein each positive sample pair comprises a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and 
 training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting a sum of at least one second distance corresponding to the negative sample pair from a sum of at least one first distance corresponding to the positive sample pair, wherein the at least one first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the at least one second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network. 
 
     
     
       10. The non-transitory, computer-readable medium of  claim 6 , wherein the distance is a Euclidean distance. 
     
     
       11. A computer-implemented system, comprising:
 one or more computers; and 
 one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform one or more operations comprising:
 obtaining a group of N vehicle accident images, wherein N is a natural number greater than or equal to 2; 
 obtaining N feature vectors respectively corresponding to the vehicle accident images by inputting the vehicle accident images into a trained convolutional neural network; 
 calculating respective distances between pairs of feature vectors among the N feature vectors, the pairs of feature vectors corresponding respectively to pairs of images among the N vehicle accident images; 
 determining that a pair of images of the N vehicle accident images corresponding to the distance are abnormal when the distance is greater than a first predetermined threshold; 
 determining a number of the vehicle accident images that are abnormal, wherein the number of the vehicle accident images that are abnormal is equal to M; 
 calculating a first probability based on M and N; 
 determining that the first probability is greater than a second predetermined threshold; and 
 in response, classifying the group of the N vehicle accident images as abnormal. 
 
 
     
     
       12. The computer-implemented system of  claim 11 , wherein the trained convolutional neural network is trained using at least one positive sample pair, wherein the training comprises:
 obtaining at least one positive sample pair, wherein each positive sample pair comprises a first image and a second image, and the first image and the second image correspond to the same vehicle accident scene; and 
 training the convolutional neural network by using the at least one positive sample pair, to decrease a sum of at least one first distance respectively corresponding to the positive sample pair, wherein the at least one first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network. 
 
     
     
       13. The computer-implemented system of  claim 11 , wherein the trained convolutional neural network is trained by at least one negative sample pair, wherein the training comprises:
 obtaining at least one negative sample pair, wherein each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and 
 training the convolutional neural network by using the at least one negative sample pair, to increase a sum of at least one second distance respectively corresponding to the negative sample pair, wherein the at least one second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network. 
 
     
     
       14. The computer-implemented system of  claim 11 , wherein the trained convolutional neural network is trained by at least one positive sample pair and at least one negative sample pair, wherein the training comprises:
 obtaining at least one positive sample pair and at least one negative sample pair, wherein each positive sample pair comprises a first image and a second image, the first image and the second image correspond to the same vehicle accident scene, each negative sample pair comprises a third image and a fourth image, and the third image and the fourth image correspond to different vehicle accident scenes; and 
 training the convolutional neural network by using the at least one positive sample pair and the at least one negative sample pair, to decrease a value obtained by subtracting a sum of at least one second distance corresponding to the negative sample pair from a sum of at least one first distance corresponding to the positive sample pair, wherein the at least one first distance is a distance between a feature vector of the first image that is obtained by using the convolutional neural network and a feature vector of the second image that is obtained by using the convolutional neural network, and the at least one second distance is a distance between a feature vector of the third image that is obtained by using the convolutional neural network and a feature vector of the fourth image that is obtained by using the convolutional neural network. 
 
     
     
       15. The computer-implemented system of  claim 11 , wherein the distance is a Euclidean distance.